YAMDA: thousandfold speedup of EM-based motif discovery using deep learning libraries and GPU

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Motif discovery in large biopolymer datasets is computationally intensive.
• Existing popular tools like MEME have quadratic time complexity, leading to long runtimes for omics data.
• There is a need for faster motif discovery tools that maintain accuracy.

Purpose of the Study:

• To introduce YAMDA, a highly scalable motif discovery software package.
• To provide a faster alternative to MEME for motif discovery in large datasets.
• To leverage deep learning for efficient and accurate motif identification.

Main Methods:

• YAMDA is built on Pytorch, a deep learning library optimized for tensor computations.
• The software utilizes GPU acceleration for enhanced performance.
• It employs algorithms with linear time complexity for motif finding.

Main Results:

• YAMDA achieves motif discovery accuracy comparable to MEME.
• The software runs in linear time, completing analyses in seconds or minutes.
• This results in speedups exceeding a thousandfold compared to traditional methods.

Conclusions:

• YAMDA offers a highly scalable and efficient solution for motif discovery.
• The software significantly reduces computational time for large biopolymer sequence datasets.
• YAMDA is freely available, facilitating its use in omics research.